Ideal algorithms in healthcare: Explainable, dynamic, precise, autonomous, fair, and reproducible

Area of Science:

• Artificial Intelligence in Healthcare
• Medical Informatics
• Algorithm Development

Background:

• Existing guidelines focus on minimum reporting standards for healthcare algorithms.
• There is a need to define characteristics of ideal algorithms for optimal patient, clinician, and investigator benefit.

Purpose of the Study:

• To propose a framework for ideal healthcare algorithms.
• To outline six desiderata for ideal algorithms: explainable, dynamic, precise, autonomous, fair, and reproducible.
• To provide a checklist for evaluating algorithms and suggest improvement strategies.

Main Methods:

• Development of a framework for ideal algorithms based on six desiderata.
• Creation of an ideal algorithms checklist.
• Application of the checklist to highly cited healthcare algorithms.

Main Results:

• The proposed framework includes explainability, dynamism, precision, autonomy, fairness, and reproducibility as key characteristics.
• The checklist can be used to assess current algorithms against ideal standards.
• Strategies like the PDR framework and SPIRIT-AI extension can guide algorithm development.

Conclusions:

• Achieving ideal healthcare algorithms requires focusing on explainability, dynamism, precision, autonomy, fairness, and reproducibility.
• The proposed framework and checklist offer practical tools for enhancing healthcare algorithms.
• Adoption of suggested strategies can lead to more beneficial and trustworthy AI in healthcare.